For a long time, the secret to more performance was to execute more instructions per cycle, otherwise known as Instruction Level Parallelism (ILP), or decreasing the latency of instructions. To execute more instructions each cycle, more functional units (e.g., integer, floating point, load/store units, etc.) have to be added. In order to more consistently execute multiple instructions, a processing paradigm called out-of-order processing (OOP) may be used, and in fact, this type of processing has become mainstream.
OOP arose because many instructions are dependent upon the outcome of other instructions, which have already been sent into the processing pipeline. To help alleviate this problem, a larger number of instructions are stored in order to allow immediate execution. The reason this is done is to find more instructions that are not dependent upon each other. The area of storage used to store the instructions that are ready to execute immediately is called the reorder buffer. The size of reorder buffers have been growing in most modern commercial computer architectures with some systems able to store as many as 126 instructions. The reason for increasing the size of the reorder buffer is simple: code that is spatially related tends also to be temporally related in terms of execution (with the possible exclusion of arrays of complex structures and linked lists). The only problem is that these instructions also have a tendency to depend upon the outcome of prior instructions. With a CPU's ever increasing amount of required code, the only current way to find more independent instructions has been to increase the size of the reorder buffer.
However, using this technique has achieved a rather impressive downturn in the rate of increased performance and in fact has been showing diminishing returns. It is now taking more and more transistors to achieve the same rate of performance increase. Instead of focusing intently upon uniprocessor ILP extraction, one can focus upon a coarser form of extracting performance at the instruction or thread level, via multithreading (multiprocessing), but without the system bus as a major constraint.
The ability to put more transistors on a single chip has allowed on-chip multiprocessing (CMP). To take advantage of the potential performance increases, the architecture cannot use these multiple processors as uniprocessors but rather must use multiprocessing that relies on executing instructions in a parallel manner. This requires the programs executed on the CMP to also be written to execute in a parallel manner rather than in a purely serial or sequential manner. Assuming that the application is written to execute in a parallel manner (multithreaded), there are inherent difficulties in making the program written in this fashion execute faster proportional to the number of added processors.
The general concept behind using multiple cores on one die is to extract more performance by executing two threads at once. By doing so, the two CPUs together are able to keep a higher percentage of the aggregate number of functional units doing useful work at all times. If a processor has more functional units, then a lower percentage of those units may be doing useful work at any one time. The on-chip multiprocessor lowers the number of functional units per processor, and distributes separate tasks (or threads) to each processor. In this way, it is able to achieve a higher throughput on both tasks combined. A comparative uniprocessor would be able to get through one thread, or task, faster than a CMP chip could, because, although there are wasted functional units, there are also “bursts” of activity produced when the processor computes multiple pieces of data at the same time and uses all available functional units. One idea behind multiprocessors is to keep the individual processors from experiencing such burst activity times and instead have each processor use what resources it has available more frequently and therefore efficiently. The non-use of some of the functional units during a clock cycle is known as “horizontal waste,” which CMP tries to avoid.
However, there are problems with CMP. The traditional CMP chip sacrifices single-thread performance in order to expedite the completion of two or more threads. In this way, a CMP chip is comparatively less flexible for general use, because if there is only one thread, an entire half of the allotted resources are idle and completely useless (just as adding another processor in a system that uses a singly threaded program is useless in a traditional multiprocessor (MP) system). One approach to making the functional units in a CMP more efficient is to use course-grained multithreading (CMT). CMT improves the efficiency with respect to the usage of the functional units by executing one thread for a certain number of clock cycles. The efficiency is improved due to a decrease in “vertical waste.” Vertical waste describes situations in which none of the functional units are working due to one thread stalling.
When switching to another thread, the processor saves the state of that thread (i.e., it saves where instructions are in the pipeline, which units are being used) and switches to another one. It does so by using multiple register sets. The advantage of this is due to the fact that often a thread can only go for so long before it falls upon a cache miss, or runs out of independent instructions to execute. A CMT processor can only execute as many different threads in this way as it has support for. So, it can only store as many threads as there are physical locations for each of these threads to store the state of their execution. An N-way CMT processor would therefore need to have the ability to store the state of N threads.
A variation on this concept would be to execute one thread until it has experienced a cache miss (usually a L2 (secondary) cache miss), at which point the system would switch to another thread. This has the advantage of simplifying the logic needed to rotate the threads through a processor, as it will simply switch to another thread as soon as the prior thread is stalled. The penalty of waiting for a requested block to be transferred back into the cache is then alleviated. This is similar to the hit under miss (or hit under multiple miss) caching scheme used by some processors, but it differs because it operates on threads instead of upon instructions. The advantages of CMT over CMP are CMT does not sacrifice single-thread performance, and there is less hardware duplication (less hardware that is halved to make the two processors “equal” to a comparable CMT).
A more aggressive approach to multithreading is called fine-grained multithreading (FMT). Like CMT, the basis of FMT is to switch rapidly between threads. Unlike CMT, however, the idea is to switch each and every cycle. While both CMT and FMT actually do indeed slow down the completion of one thread, FMT expedites the completion of all the threads being worked on, and it is overall throughput which generally matters most.
CMPs may remove some horizontal waste in and unto themselves. CMT and FMT may remove some (or all) vertical waste. However an architecture that comprises an advanced form of multithreading, referred to as Simultaneous Multithreading (SMT), may be used to reduce both horizontal and vertical waste. The major goal of SMT is to have the ability to run instructions from different threads at any given time and in any given functional unit. By rotating through threads, an SMT architecture acts like an FMT processor, and by executing instructions from different threads at the same time, it acts like CMP. Because of this, it allows architects to design wider cores without the worry of diminishing returns. It is reasonable for SMT to achieve higher efficiency than FMT due to its ability to share “unused” functional units among differing threads; in this way, SMT achieves the efficiency of a CMP machine. However, unlike a CMP system, an SMT system makes little to no sacrifice (the small sacrifice is discussed later) for single threaded performance. The reason for this is simple. Whereas much of a CMP processor remains idle when running a single thread and the more processors on the CMP chip makes this problem more pronounced, an SMT processor can dedicate all functional units to the single thread. While this is obviously not as valuable as being able to run multiple threads, the ability to balance between single thread and multithreaded environments is a very useful feature. This means that an SMT processor may exploit thread-level parallelism (TLP) if it is present, and if not, will give full attention to instruction level parallelism (ILP).
In order to support multiple threads, an SMT processor requires more registers than the traditional superscalar processor. The general aim is to provide as many registers for each supported thread as there would be for a uniprocessor. For a traditional reduced instruction set computer (RISC) chip, this implies 32 times N registers (where N is the number of threads an SMT processor could handle in one cycle), plus whatever renaming registers are required. For a 4-way SMT processor RISC processor, this would mean 128 registers, plus however many renaming registers are needed.
Most SMT models are straightforward extensions of a conventional out-of-order processor. With an increase in the actual throughput comes more demands upon instruction issue width, which should be increased accordingly. Because of the aforementioned increase in the register file size, an SMT pipeline length may be increased by two stages (one to select register bank and one to do a read or write) so as not to slow down the length of the clock cycle. The register read and register write stages are therefore both broken up into two pipelined stages.
In order to not allow any one thread to dominate the pipeline, an effort should be made to ensure that the other threads get a realistic slice of the execution time and resources. When the functional units are requesting work to do, the fetch mechanism will provide a higher priority to those threads that have the fewest instructions already in the pipeline. Of course, if the other threads have little they can do, more instructions from the thread are already dominating the pipelines.
SMT is about sharing whatever possible. However, in some instances, this disrupts the traditional organization of data, as well as instruction flow. The branch prediction unit becomes less effective when shared, because it has to keep track of more threads with more instructions and will therefore be less efficient at giving an accurate prediction. This means that the pipeline will need to be flushed more often due to miss prediction, but the ability to run multiple threads more than makes up for this deficit.
The penalty for a misprediction is greater due to the longer pipeline used by an SMT architecture (by two stages), which is in turn due to the rather large register file required. However, techniques have been developed to minimize the number of registers needed per thread in an SMT architecture. This is done by more efficient operating system (OS) and hardware support for better deallocation of registers, and the ability to share registers from another thread context if another thread is not using all of them.
Another issue is the number of threads in relation to the size of caches, the line sizes of caches, and the bandwidth afforded by them. As is the case for single-threaded programs, increasing the cache-line size decreases the miss rate but also increases the miss penalty. Having support for more threads which use more differing data exacerbates this problem and thus less of the cache is effectively useful for each thread. This contention for the cache is even more pronounced when dealing with a multiprogrammed workload over a multithreaded workload. Thus, if more threads are in use, then the caches should be larger. This also applies to CMP processors with shared L2 caches.
The more threads that are in use results in a higher overall performance and the differences in association of memory data become more readily apparent. There is an indication that when the L1 (primary) cache size is kept constant, the highest level of performance is achieved using a more associative cache, despite longer access times. Tests have been conducted to determine performance with varying block sizes that differ associatively while varying the numbers of threads. As before, increasing the associative level of blocks increased the performance at all times; however, increasing the block size decreased performance if more than two threads were in use. This was so much so that the increase in the degree of association of blocks could not make up for the deficit caused by the greater miss penalty of the larger block size.
If a thread enters the code stream that is not critical to system performance, it should be assigned the lowest possible priority by the software. This practice may be used to conserve power since the hardware performance would not be a concern. In a high-performance SMT processor, power consumption is always a concern. There is, therefore, a need for a method that conserves power based on the execution priority of instructions in the multiple threads.